Profiling of the Tox21 10K compound library for agonists and antagonists of the estrogen receptor alpha signaling pathway

The U.S. Tox21 program has screened a library of approximately 10,000 (10K) environmental chemicals and drugs in three independent runs for estrogen receptor alpha (ERα) agonist and antagonist activity using two types of ER reporter gene cell lines, one with an endogenous full length ERα (ER-luc; BG1 cell line) and the other with a transfected partial receptor consisting of the ligand binding domain (ER-bla; ERα β-lactamase cell line), in a quantitative high-throughput screening (qHTS) format. The ability of the two assays to correctly identify ERα agonists and antagonists was evaluated using a set of 39 reference compounds with known ERα activity. Although both assays demonstrated adequate (i.e. >80%) predictivity, the ER-luc assay was more sensitive and the ER-bla assay more specific. The qHTS assay results were compared with results from previously published ERα binding assay data and showed >80% consistency. Actives identified from both the ER-bla and ER-luc assays were analyzed for structure-activity relationships (SARs) revealing known and potentially novel ERα active structure classes. The results demonstrate the feasibility of qHTS to identify environmental chemicals with the potential to interact with the ERα signaling pathway and the two different assay formats improve the confidence in correctly identifying these chemicals.


Materials and Methods
qHTS of ERα beta-lactamase and BG1 ER luciferase reporter gene assays. The HEK293 ER-bla cells were cultured in assay medium (DMEM phenol red free medium containing 2% charcoal stripped FBS) overnight and the BG1 ER-Luc cells were in assay medium (DMEM phenol red free medium containing 10% charcoal stripped FBS) for 5 days prior to the screening. Both agonist and antagonist screen protocols were summarized in Table   S1 and Table S2. were incubated at 37°C for 22 hr. After 5 µL of ONE-Glo™ Luciferase Assay reagent (Promega, Madison, WI, USA) was added using an FRD (Aurora Discovery), the plates were incubated at RT for 30 min, and luminescence intensity was measured by ViewLux plate reader (Perkin Elmer). Data was expressed as the ratio of 460 nm/530 nm emissions for HEK293 ER-bla assay and expressed as relative luminescence units for BG1 ER-Luc assay. For primary data analysis, raw plate reads for each titration point were first normalized relative to β-estradiol control (10 nM for the BG1 ER-Luc assay and 40 nM for the HEK293 ER-bla assay, 100%) and DMSO only wells (basal, 0%).
For antagonist mode screening, the assays were run multiplexed with a cell viability assay, HEK293 ER-bla cells suspended in assay medium were dispensed at 5,000 cells/5 µL/well in 1536-well black wall/clear bottom plates (Greiner Bio-One) and BG1Luc4E2 cells dispensed at 4,000 BG1 ER-Luc cells/4 µL/well in 1536-well white tissue cultured plates (Greiner Bio-One) using a Thermo Scientific Multidrop Combi (Thermo Fisher Scientific Inc.). After assay plates were incubated at 37ºC for 5 h (HEK293 ER-bla assay) or 24 h (BG1 ER-Luc assay), 23 nL of compounds dissolved in DMSO, positive controls, or DMSO was transferred to the assay plate by a pintool (Kalypsys), followed by the For HEK293 ER-bla assay, 1 µL of LiveBLAzer™ B/G FRET substrate (Invitrogen) was added into each well using an FRD Dispenser (Aurora Discovery). After the assay plates were incubated for 2 h at RT, the fluorescence intensity at 460 and 530 nm emissions was measured at 405 nm excitation by an Envision plate reader (Perkin Elmer), followed by an addition of 4 µL/well of cell viability reagent (CellTiter-Glo, Promega) using a FRD (Aurora Discovery). After 30 min incubation at RT, the luminescence intensity in the plates was measured using a ViewLux plate reader (Perkin Elmer). For BG1 ER-Luc assay, 1 µL/well of CellTiter-Fluor reagent (Promega) was added into the assay plates using an FRD (Aurora Discovery). After 30 min incubation at 37 o C, the fluorescence intensity in the plates was measured using a ViewLux plate reader (PerkinElmer), followed by the addition of 4 μL of ONE-Glo™ Luciferase Assay reagent (Promega) using an FRD (Aurora Discovery), the plates were incubated at RT for 30 min, and luminescence intensity was measured by ViewLux plate reader (Perkin Elmer). Data was expressed as the ratio of 460 nm/530 nm emissions for HEK293 ER-bla assay and expressed as relative luminescence units for BG1Luc4E2 assay. For primary data analysis, raw plate reads for each titration point were first normalized relative to 0.5 nM β-estradiol in presence of 1 µM 4-hydroxy tamoxifen (ER antagonist mode) or tetra n-octyl ammonium bromide (cell viability) control (-100%) and 0.5 nM β-estradiol (control, 0%).
Auto-fluorescence assay. The auto-fluorescence assay was performed in the cells or cell free assay medium. The cells or assay medium only were dispensed at 2,000 cells/5 µL/well or at 5 µL assay medium/well in 1536-well black wall/clear bottom plates (Greiner Bio-One) using a Scientific Multidrop Combi (Thermo Fisher Scientific Inc.).
After the assay plates were incubated at a 37ºC/5% CO2 incubator for 5 hr, 23   Activity assignments based on triplicate run. Each curve class was first converted to a curve rank as previously described 1 such that more potent and efficacious compounds with higher quality curves were assigned a higher rank. Curve ranks should be viewed as qualitative descriptors of the concentration response activity of the compound. Curve ranks from replicate runs of a compound were averaged, and the activity outcome of each compound in the BG1 ER-luc assays and from each readout of the HEK293 ER-bla assays (ratio, 460 nm, 530 nm and cell viability) was assigned based on its average curve rank and reproducibility call as shown in Table S3. The final activity outcome of each compound was determined based on its multi channel readout activity as shown in Table   S4. For antagonist mode assays, the cell viability counter screen data were used to flag potential cytotoxic artifacts. For the HEK293 ER-bla agonist mode assay, potential artifacts produced by blue fluorescent compounds were flagged using both the compound auto fluorescence profiling data and promiscuous compound activity shown in the 460 nm readout of all the additional BLA assays screened in Tox21 with this compound library.

Identification of auto fluorescence (HEK293 ER-bla agonist mode assay) and cytotoxicity
(antagonist mode ER assays) artifacts. In the HEK293 ER-bla agonist mode assay, blue fluorescent compounds could show the same phenotype as agonists. Two approaches were used to identify potential auto fluorescent artifacts to distinguish them from true ER agonists. One approach is the auto fluorescence detection counter screen measured at 460 nm (blue) in the ER background HEK293 cells and cell free medium. Any sample with the agonist phenotype in the blue channel of the BLA assay that also showed activation in the auto fluorescence counter screen with an AC50 difference <3 fold was identified as a potential auto fluorescent false positive. This approach identified 34 such samples. The second approach is examining the activity of each sample in all the eight BLA assays screened to date in Tox21 10K library. Samples with the agonist phenotype in the blue channel of the HEK293 ER-bla assay that also had an >4 average curve rank in the blue channel of all the BLA assays were considered promiscuously active in BLA assays and potentially blue fluorescent. This second approach identified 39 potential fluorescent false positives, 24 of which overlapped with those identified by the first approach. The two approaches yielded a total of 49 unique samples that were assigned the "inconclusive agonist (fluorescent)" activity outcome category, most of which are known blue fluorescent compounds (see PubChem 4 assay ID 743077).
The BG1 ER-luc assay has a luminescence-based readout so that interference from compound auto fluorescence was not a concern for this assay. The activity outcomes of the 49 samples identified as auto fluorescent in the HEK293 ER-bla assay were examined in the BG1 ER-luc assay and 19 of them were assigned the "active agonist" category, the ER activity of which could be, therefore, real and not artificial, and the rest of the 49 samples were either inactive or inconclusive in the BG1 ER-luc assay.
As cytotoxic compounds could show the same inhibitory phenotype as antagonists in these assays, we need an effective strategy to distinguish true ER antagonists from a cytotoxicity-related false positive response. For this reason, each antagonist mode assay was accompanied with a cell viability readout that serves as the counter screen. As an alternative to the cell viability counter screen, the control channel (530 nm readout) of the BLA assay could be used to identify potential cytotoxic compounds. Either activation or inhibition shown in this channel could be an indication of cytotoxicity 1 . To compare the effectiveness of the 530 nm readout and the cell viability counter screen at identifying potential cytotoxicity artifacts, activity outcomes were assigned to the screened samples using both approaches (Table S4(b)) and the set of known ER reference compounds was used to evaluate the accuracy of these assignments. Both approaches achieved 100% specificity in correctly identifying the reference compounds. Filtering with the cell viability counter screen resulted in better sensitivity compared to the 530 nm readout for both all reference compounds (59% vs. 50%) and when only the known ER antagonists were used (100% vs. 70%) in the evaluation. Three known ER antagonists were misclassified as cytotoxicity artifacts using the 530 nm readout resulting in lower sensitivity.

Structure classes of identified ER agonists and antagonists -partial vs. full-length
receptor cell lines. Other hydroxylated aromatic hydrocarbons that co-cluster with the phenols have been reported to show estrogenic activities, such as 2-and 3hydroxyfluorenes and 1-hydroxypyrene 5 , which are commonly found in cigarette smoke condensate 6 and identified as active agonists in both the HEK293 ER-bla and BG1 ER-luc assays. Alkylphenols, such as nonylphenols, are precursors to commercially important detergents and are produced in large volumes annually 7 . Nonylphenols and other industrial alkylphenols have been known as synthetic environmental estrogens with generally weak estrogenic activities reported [8][9][10] . The cluster of alkylphenols, including nonylphenols, in the 10K library was found enriched with active agonists in both of the ER agonist mode assays with some compounds also acting as antagonists in the ER antagonist mode assays.
Polycyclic aromatic hydrocarbons (PAHs) are a large group of environmental and dietary toxicants, which and/or the metabolites of which have been reported to show either estrogenic or antiestrogenic activities [11][12][13][14] . Our cluster of PAHs showed an enrichment of active agonists in the ER agonist mode assays (more so in the HEK293 ER-bla assay) and enrichment of active antagonists in the BG1 ER-luc antagonist mode assay. Of particular note was the PAH 7,12-dimethylbenz(a)anthracene, which acted as an agonist in the BG1 ER-luc agonist mode assay and as an antagonist in the corresponding antagonist mode assay at low concentrations (<1 µM) but started to act like an agonist at higher concentrations. These compounds are reported to require reduction by cytochrome P450 monooxygenase to generate phenolic groups. These compounds are usually reported as negative in ER binding assays lacking metabolism. Activity in the assays reported here may indicate some metabolic capacity of the cell lines used. A similar class of compounds, the polycyclic quinones and phenyl ketones 15 , on the other hand, showed a significant enrichment of active agonists in the BG1 ER-luc assay in both agonist and antagonist modes, but less so in the HEK293 ER-bla assay. No ER activity has been reported previously for some compounds in this class (e.g., dibenzosuberone, which was active in both the BG1 ER-luc and HEK293 ER-bla assays). Organophosphates commonly used as flame retardants is another class of compounds that only showed enrichment in active agonists in the BG1 ER-luc assay and not the HEK293 ER-bla assay. The ER activities of some of these compounds (e.g., triphenyl phosphate), which was also identified by the HEK293 ER-bla assay as an inconclusive agonist, have been reported recently 16 .
The trimethylolpropane trimethacrylate class of compounds is a monomer used in the manufacture of acrylics and plastic components in a wide variety of products. These chemicals used as dental materials have been tested for estrogenic activity in vitro but with negative results 17 . This acrylate class of compounds, however, showed an enrichment of active agonists in our HEK293 ER-bla agonist mode assay. These apparent ER agonists include acrylates such as trimethylolpropane triacrylate, pentaerythritol triacrylate, dipentaerythritol pentaacrylate, and pentaerythritol tetraacrylate. These compounds also acted as agonists in the BG1 ER-luc antagonist mode at concentrations <30 µM, but the signals started to drop at higher concentrations perhaps due to cytotoxicity. Some of these compounds acted in a similar fashion in the HEK293 ER-bla antagonist mode and BG1 ER-luc agonist mode assays as well with some ER activation shown at low concentrations and inhibition due to cytotoxicity at high concentrations.
There is a group of organometalic compounds in the 10K library including triphenyltin, triphenyllead, triphenylbismuth, and phenylmercuric compounds that showed an enrichment of active agonists in the HEK293 ER-bla agonist mode assay.
Triphenylbismuth also acted as an agonist in the BG1 ER-luc assay. Organotin compounds have recently been reported to inhibit the transcriptional activation of human ER 18 . No report has been found on the estrogenic activity of the other organometalic compounds. These organometalic compounds showed inhibition activity in the antagonist mode HEK293 ER-bla assay and the BG1 ER-luc assays but were mostly classified as inconclusive antagonists due to cytotoxicity. Although no clear indication of ER activity has been reported for the class of cinnamates, commonly used in sunscreens and cosmetics 19 , this structure class showed an enrichment of active agonists only in the BG1 ER-luc assay. Some of the cinnamates, such as 3-methylbutyl cinnamate and 2phenylethyl 3-phenylprop-2-enoate, acted as agonists in the antagonist mode BG1 ER-luc assay as well. Negative to weak ER binding activities have been reported for compounds belong to the class of phenyl benzoates including phenyl parabens, phenyl phthalates, and phenyl salicylates, which are also common ingredients of cosmetics and sunscreens 20,21 .
Similar to the cinnamates, we found this class of compounds enriched with active agonists in the BG1 ER-luc assay with most showing agonist activity also in the antagonist mode assay. Phenyl 4-aminosalicylate and phenylparaben were also found active in the HEK293 ER-bla agonist mode assay.
The cluster of chloranocryl herbicides was also enriched with active antagonists (e.g., cypromid, chloranocryl, propanil) only in the BG1 ER-luc assay. Of these herbicides, only propanil has been previously reported to show endocrine disruption but not through direct ERα binding 22 . No previous report of ER activity has been found on the other herbicides in this cluster. The well known histone deacetylase (HDAC) inhibitor, suberoylanilide hydroxamic acid (SAHA or vorinostat), was closely clustered with the chloranocryl herbicides and identified as an active antagonist in the BG1 ER-luc assay.
SAHA has been reported to induce ERα degradation in the breast cancer MCF-7 cells via the C-terminal Hsc70 interacting protein-mediated ubiquitin pathway [23][24][25][26] , which may explain its ER antagonist activity observed. The retinoic acids acted as active antagonists in BG1 ER-luc assay, but as active agonists in HEK293 ER-bla assay. Retinoic acid has been found previously to inhibit ER activity not through direct ER-binding but by altering the amount of ER protein bound to the ERE or affecting the transcriptional efficiency of this complex 27 . Another study found that retinoic acids decreased ERα expression in estrogen-responsive endometrial cancer cells 28 . The phenyl carboxamides are in one of the neighboring clusters of the retinoic acids. This cluster was enriched with active antagonists in the BG1 ER-luc assay but also contain some active agonists. The active antagonists in this class of compounds include retinoic acid receptor subtypespecific ligands, retinoid analogs AM580 29,30 and tamibarotene (Am80) 31 . No specific ER activity has been reported previously on these compounds but the structure activity analyses suggest that they may act in a fashion similar to the retinoic acids. The cluster of triazole fungicides was another structure class enriched in active antagonists (e.g., metconazole, diniconazole, penconazole, tebuconazole, myclobutanil, propiconazole, epoxiconazole, hexaconazole, fenbuconazole, tetraconazole, ipconazole) in the BG1 ERluc assay with fewer actives in the HEK293 ER-bla assay. Of these compounds, only myclobutanil has been reported to bind to ERα and have antiestrogenic effect 32 . The benzodiazepine class of psychoactive drugs was also found enriched with active antagonists in the BG1 ER-luc assay and vatalanib, a tyrosine kinase inhibitor closely clustered with the benzodiazepines, was identified as an active antagonist in both the BG1 ER-luc and the HEK293 ER-bla assay. However, none of these compounds has been reported previously to have ER activity.
Artemisinin and its derivatives are used as standard treatments for malaria 33 . The class of artemisinin analogs was also one of the classes found enriched with active antagonists only in the HEK293 ER-bla assay. Consistent with this observed antiestrogenic activity, artemisinin has been reported to selectively decrease functional levels of ERα and ablate estrogen-induced proliferation in human breast cancer cells 34 . No direct ER binding was reported, however. The class of DNA intercalating agents, enriched with active antagonists in the HEK293 ER-bla assay, showed interesting activities in both the HEK293 ER-bla and BG1 ER-luc assays. Rubitecan and actinomycin D were identified as active antagonists in both HEK293 ER-bla and BG1 ER-luc. Other compounds, such as daunorubicin, idarubicin, carminomycin, aclarubicin, plicamycin, chromomycin A3, adriamycin, and daunomycin, all acted as active antagonists in the HEK293 ER-bla assay, but in the BG1 ER-luc assay, some acted as antagonists and others acted as agonists. The only previous report related to their ER activity investigated actinomycin D, which was found to prevent the nuclear processing of ER 35 . A more recent report related to the ER activity of DNA intercalators was about XR5944, a compound not in the Tox21 10K library. XR5944 was reported to specifically inhibit the binding of ER to its consensus DNA sequence and its subsequent activity 36 . The DNA intercalating agents with apparent ER antagonist activity in our ER assays are likely to exert their activity in a similar fashion. Vinca alkaloids, including the antimicrotubule agents vinblastine and vincristine, were also found enriched with active antagonists in HEK293 ER-bla assay and less so in BG1 ER-luc assay. Vinca alkaloids have been reported to decrease ERα protein levels in the human breast cancer cell line MCF-7 and inhibit estradiol mediated transactivation at ERE-driven promoters 37 . Again, no direct ER binding was indicated.
Finally, the class of glycol acrylates, which has wide industrial applications such as adhesives, solvents, coating materials and cosmetics, was found enriched with active antagonists in HEK293 ER-bla assay with some compounds acted as active agonists (e.g., ethylene acrylate) in both ER assays. However, no previous report has been found on this class of compounds exhibiting ER activity.
Functional assay vs. binding assay. Many steroid hormones, including testosterone, progesterone and their analogs, showed no affinity to ER in the binding assay but acted as agonists in the BG1 ER-luc assay or both reporter gene assays. Testosterone showed a 15 nM potency in the BG1 ER-luc assay and 10 µM in the HEK293 ER-bla assay.
Androgens have been reported to activate the translocation of ER and induce the synthesis of the uterine-induced protein with barely detectable affinity to the cytosol ER 38,39 . Progesterone was identified as an active agonist in the BG1 ER-luc agonist mode assay with 4 µM potency, an active antagonist in the BG1 ER-luc antagonist mode assay with 58 µM potency, and was inactive in the HEK293 ER-bla assay. Progesterone has been reported to antagonize estrogen action not through interactions with ER but via progestin receptors 40,41 . Progesterone thus may require a full length receptor to exert its activity on ER. This would also explain its inactivity in the HEK293 ER-bla assay, which has only a partial receptor.
The retinoic acids (trans-retinoic acid and 13-cis retinoic acid) also showed no detectable affinity to ER in the binding assay, but acted as antagonists in the BG1 ER-luc assay and agonists in HEK293 ER-bla assay. Retinoic acid has been found previously to inhibit ER activity not through direct ER-binding but by altering the amount of ER protein bound to the ERE or affecting the transcriptional efficiency of this complex 27 . Another study in addition found that retinoic acids decreased ERα expression in estrogen-responsive endometrial cancer cells through crosstalk with the estrogen signaling pathway 28 .   Step Notes       are colored yellow, non-binders are colored black, and compounds with no binding data are colored gray.